Speex dev - Mar 2008 - Speex requirements on a TI Davinci / ARM926EJ-Sid(wb)

(Resending since the archive seems to have eaten the text.)

 

I'm working on getting Speex running on a TI Davinci (6441), on both the
ARM and the DSP. My immediate goal is to understand the processing
requirements on different architectures so that I can properly specific
an engineering project.

 

I have it up and running on the ARM, and I just wanted to see if anyone
could sanity-check my results before I continue.

 

Brief version:

WB decode takes ~24MIPS, encode takes ~243MIPS.

NB decode takes ~10MIPS, encode takes ~102MIPS.

(And by MIPS, I mean ARM CPU cycles, not Dhrystone MIPS)

 

If I manually (because I couldn't figure out how to make configure do
it) add "-mpcu=arm926ej_s" and "-DSHORTCUTS", the wideband
numbers drop
to 22MIPS and 219MIPS, respectively.

 

More Details:

 

The commands/options I used to config Speex:

 

# export ARM_INSTALL_TREE=/extern/toddk/workdir/filesys

export CC=arm_v5t_le-gcc

export CXX=arm_v5t_le-g++

export AR=arm_v5t_le-ar

export RANLIB=arm_v5t_le-ranlib

export NM=arm_v5t_le-nm

export ac_cv_path_NM=arm_v5t_le-nm

export ac_cv_func_setpgrp_void=yes

export host=arm_v5t_le

 

./configure --prefix /extern/toddk/workdir/filesys/usr/local
--host=arm-linux --disable-float-api --disable-vbr --enable-fixed-point
--enable-arm5e-asm

 

The source data is from MP3 audio books that have been resampled to 16K
and written out as .WAV files by NCH's Switch software. (It's hard to
get good wideband audio samples...)

 

For decode, I run the command "time speexdec sample.spx", and then
take
sum of the "user" and "sys" times the speed of the processor
(297MHz),
and divide it by the "real" (or clock) time. That should give me the
average processing power requirements. (And the number is matched by the
results of "top".)

 

For encode, I run the command "time speexenc sample.wav - >
/dev/null".
Since it consumes 99% of the cpu, and runs faster than real time, the
MIPS requirement is calculated by taking sum of the "user" and
"sys"
times the speed of the processor (297MHz), and dividing by the playback
time determined by the decode step.

 

Does this seem like a reasonable way to test the code?

Are the results in line with what people would expect?

 

Thanks!

---------------------------

Todd Krein

Director of Architecture

650-566-6657

todd.krein@ooma.com

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/speex-dev/attachments/20080303/99db6c43/attachment.html

Did you try setting complexity to 1 (defaults to 2)?

Todd Krein a ?crit :
> (Resending since the archive seems to have eaten the text.)
> 
>  
> 
> I?m working on getting Speex running on a TI Davinci (6441), on both the
> ARM and the DSP. My immediate goal is to understand the processing
> requirements on different architectures so that I can properly specific
> an engineering project.
> 
>  
> 
> I have it up and running on the ARM, and I just wanted to see if anyone
> could sanity-check my results before I continue.
> 
>  
> 
> Brief version:
> 
> WB decode takes ~24MIPS, encode takes ~243MIPS.
> 
> NB decode takes ~10MIPS, encode takes ~102MIPS.
> 
> (And by MIPS, I mean ARM CPU cycles, not Dhrystone MIPS)
> 
>  
> 
> If I manually (because I couldn?t figure out how to make configure do
> it) add ?-mpcu=arm926ej_s? and ?-DSHORTCUTS?, the wideband numbers drop
> to 22MIPS and 219MIPS, respectively.
> 
>  
> 
> More Details:
> 
>  
> 
> The commands/options I used to config Speex:
> 
>  
> 
> # export ARM_INSTALL_TREE=/extern/toddk/workdir/filesys
> 
> export CC=arm_v5t_le-gcc
> 
> export CXX=arm_v5t_le-g++
> 
> export AR=arm_v5t_le-ar
> 
> export RANLIB=arm_v5t_le-ranlib
> 
> export NM=arm_v5t_le-nm
> 
> export ac_cv_path_NM=arm_v5t_le-nm
> 
> export ac_cv_func_setpgrp_void=yes
> 
> export host=arm_v5t_le
> 
>  
> 
> ./configure --prefix /extern/toddk/workdir/filesys/usr/local
> --host=arm-linux --disable-float-api --disable-vbr --enable-fixed-point
> --enable-arm5e-asm
> 
>  
> 
> The source data is from MP3 audio books that have been resampled to 16K
> and written out as .WAV files by NCH?s Switch software. (It?s hard to
> get good wideband audio samples?)
> 
>  
> 
> For decode, I run the command ?time speexdec sample.spx?, and then take
> sum of the ?user? and ?sys? times the speed of the processor (297MHz),
> and divide it by the ?real? (or clock) time. That should give me the
> average processing power requirements. (And the number is matched by the
> results of ?top?.)
> 
>  
> 
> For encode, I run the command ?time speexenc sample.wav - > /dev/null?.
> Since it consumes 99% of the cpu, and runs faster than real time, the
> MIPS requirement is calculated by taking sum of the ?user? and ?sys?
> times the speed of the processor (297MHz), and dividing by the playback
> time determined by the decode step.
> 
>  
> 
> Does this seem like a reasonable way to test the code?
> 
> Are the results in line with what people would expect?
> 
>  
> 
> Thanks!
> 
> ---------------------------
> 
> Todd Krein
> 
> Director of Architecture
> 
> 650-566-6657
> 
> todd.krein@ooma.com <mailto:todd.krein@ooma.com>
> 
>  
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Speex-dev mailing list
> Speex-dev@xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev